MSMixer: Learned Multi-Scale Temporal Mixing with Complementary Linear Shortcut for Long-Term Time Series Forecasting

2026-05-04

AI Summary

This work addresses the challenge that existing lightweight MLP-based models struggle to simultaneously capture high-frequency oscillations, mid-range periodic patterns, and long-term trends in time series. To this end, the authors propose MSMixer, a channel-independent multi-scale MLP architecture that processes features at 1×, 4×, and 16× downsampled temporal resolutions in parallel. These multi-scale representations are dynamically fused via a learnable Softmax gating mechanism, while a DLinear shortcut pathway preserves global trend and seasonal components. The model achieves linear complexity O(T) with only 112K parameters (for prediction horizon H=96). On four ETT benchmarks, it attains an average MSE of 0.357, outperforming DLinear and NLinear in 12 out of 16 settings. Moreover, it matches or exceeds the performance of Transformer-based models in 9 tasks while using just one-fifth the parameters of PatchTST.

Abstract

Long-term time series forecasting requires models that simultaneously capture rapid oscillations, medium-range periodicities, and slowly evolving macro-trends from a fixed look-back window. Existing lightweight MLP-based models typically operate on a single temporal resolution, limiting their ability to explicitly model patterns at multiple scales. We propose MSMixer, a channel-independent multi-scale MLP architecture that addresses this limitation through three complementary innovations: (i) three parallel scale branches at down-sample factors {1x, 4x, 16x} with independent MLP blocks, (ii) a learnable softmax gate that dynamically weighs branch outputs, and (iii) a DLinear complementary shortcut that provides full-window trend and seasonality context. MSMixer contains only 112K parameters at H=96 and runs at O(T) complexity. Evaluated on four ETT benchmarks with standard chronological splits and three random seeds, MSMixer achieves the lowest average MSE (0.357) among lightweight models, outperforming DLinear (0.386, -7.4%) and NLinear (0.365, -2.1%), winning 12 of 16 configurations. Against five Transformer-based baselines from the literature, MSMixer achieves best or second-best MSE in 9 of 16 configurations while using 5x fewer parameters than PatchTST. Ablation and sensitivity analyses confirm the complementary contributions of the multi-scale branches and the DLinear shortcut.

Problem

• long-term time series forecasting
• multi-scale temporal patterns
• lightweight MLP models
• temporal resolution
• time series modeling

Innovation

• multi-scale
• MLP-based forecasting
• learnable gating
• complementary shortcut
• long-term time series